9th Conference on Limestone
Hydrogeology in Besançon
9ième Colloque d'Hydrogéologie en
Pays Calcaire à Besançon
Session 4
Metrology and data transmission
“Artificial neural network for karst
aquifer sustainable management”
Marc Lambert
Manager of Syndicat des Eaux du Vivier
9th Conference on Limestone Hydrogeology in Besançon
The Syndicat des Eaux du Vivier
« SEV » in a few figures
– 5 municipalities including the city of Niort
– about 100.000 people depending on the
Vivier karstic spring
– 650 km water networks, 36000
compteurs, 5 main water resources
– 60 agents, in public management
– Hospitals, industry with SEVESO sites,
schools…)
2
9th Conference on Limestone Hydrogeology in Besançon
What are the mains problems?
– Agricultural area with intensive
production
– Competition with water supply
– Limited and fragile underground
resources, with 2 observed karstic
collapses in Vivier spring
– Quality problems due to nitrates and
pesticides
3
9th Conference on Limestone Hydrogeology in Besançon
Our main purposes about quantity
– Securize water supply on our area
– Understand and anticipate needs
(peaks, periodicity)
– Understand and modelize water
resources (rain effect, cycles, pumping
effects…)
– Criterion:
« water resources always
> peak needs »
4
9th Conference on Limestone Hydrogeology in Besançon
Geological context
5
9th Conference on Limestone Hydrogeology in Besançon
Geological context (AB cut)
6
9th Conference on Limestone Hydrogeology in Besançon
Vivier karst observed behaviour
7
9th Conference on Limestone Hydrogeology in Besançon
Effects of pumpings for
irrigation and water supply
IMPACT DES PRELEVEMENTS AGRICOLES SUR LA RESSOURCE DE NIORT
canal
3.00
1.00
0.90
0.80
0.70
1.00
0.60
0.00
oct-01
0.50
avr-02
oct-02
avr-03
oct-03
avr-04
-1.00
oct-04
0.40
0.30
PRELEVEMENTS
(AEP/AGRI) m3/s
Débit Source (m3/s)
/Niv Puits (-m)
2.00
rabattement dans le
puits du Vivier (m)
besoin Niort (V+G)
(m3/s moyens
mensuels)
0.20
-2.00
0.10
-3.00
0.00
SEMESTRES SECS / DE RECHARGE
Irrigation (m3/s
moyens mensuels)
données estimées
pour secteur 13 8
9th Conference on Limestone Hydrogeology in Besançon
Methodology followed
– 1 : Data collecting, selecting…and mining
(ACP, geostatistics, wavelets…)
– 2 : Modelling with different techniques
(lumped modelling of rainfall – runoff
and rainfall aquifer level, inverse
modelling using fourier analysis, neural
networks…)
– 3: Forecasting with rainfall simulation
– 4: Managing resources…ie how to keep
« water resources always
needs »
> peak
9
9th Conference on Limestone Hydrogeology in Besançon
Rainfall data analysis
10
9th Conference on Limestone Hydrogeology in Besançon
Inverse modelling (TEMPO BRGM)
11
9th Conference on Limestone Hydrogeology in Besançon
Inverse modelling (TEMPO BRGM)
12
9th Conference on Limestone Hydrogeology in Besançon
Hydrological modelling
(GARDENIA BRGM)
Surface du bassin versant
PLUIE
ETP
RUMAX déficit maximal du sol
PLUIE EFFICACE
RUIPER Hauteur d'équi
Ruissellement Percolation
THG Temps ½ percolation
Débit Q
Niveau Piézom. NP
Q = SURF * EC + Qo
Percolation
(RECHARGE de la nappe)
NP = G / S + NB
Écoulement total
(DEBIT du cours d'eau)
13
9th Conference on Limestone Hydrogeology in Besançon
Hydrological modelling
(GARDENIA BRGM)
14
9th Conference on Limestone Hydrogeology in Besançon
Artificial neural network modelling
(Neuro one Netral)
15
9th Conference on Limestone Hydrogeology in Besançon
Artificial neural network modelling
16
9th Conference on Limestone Hydrogeology in Besançon
Example of use of this model:
Crisis anticipation and management
80000
0.5
70000
0
60000
-0.5
50000
-1
40000
-1.5
30000
-2
20000
-2.5
10000
-3
0
01/01/05
Ress.princ.
V+GI+GIII+Canal
(m3/j)
rabattement (<0) et pluie (>0) en m
volumes en m3/j
CROISEMENT ENTRE RESSOURCES
DISPONIBLES ET BESOINS - ETIAGE 2005
Consommation
m3/j
Secours
nécessaire m3/j
pluies (m)
rabattement
Vivier (m)
-3.5
01/07/05
01/01/06
17
9th Conference on Limestone Hydrogeology in Besançon
Conclusion: advantages of ANN tools
– After choosing a good model, easy
qnalytical formulation under Excel…
– « Intelligent » black box, but to use
under constraints (learning sets must
include extrem years)
– Easy to use for simulations (pumpings
for irrigation or water supply, climate…)
– Easy to use for sensitivity analysis
18